Nvidia acquires Kumo AI predictive model maker
Nvidia has acquired Kumo AI, a four-year-old startup that builds predictive foundation models for business data, according to Fortune and SiliconANGLE. Fortune reports the three Kumo cofounders - Vanja Josifovski, Hema Raghavan, and Stanford professor Jure Leskovec - moved to Nvidia, and that Nvidia declined to comment; The Information reported the deal was for at least $400 million. SiliconANGLE describes Kumo's platform as using graph neural networks and a proprietary Predictive Query Language, with integrations to data warehouses such as Snowflake and Databricks, and reports it automates data preparation to cut manual setup effort by as much as 95%. Fortune reports Kumo raised $37 million in 2022 from investors including Sequoia Capital, and that customers have included Reddit and Sainsbury's. The deal extends Nvidia's push into enterprise predictive AI built on relational, warehouse-connected data.
What happened
Nvidia has acquired Kumo AI, a four-year-old startup that builds predictive foundation models for business data, according to reporting by Fortune and SiliconANGLE. Fortune reports the three cofounders - Vanja Josifovski, Hema Raghavan, and Stanford professor Jure Leskovec - have moved to Nvidia, and that Nvidia declined to comment. The Information reported the deal was for at least $400 million. Fortune reports Kumo previously raised $37 million in 2022 from investors including Sequoia Capital, and that its customers have included Reddit and Sainsbury's.
Technical details
SiliconANGLE reports Kumo's models use graph neural networks to represent rows and entities in business data as nodes and edges rather than treating records in isolation, which lets the system encode relationships across customers, products, and transactions. SiliconANGLE describes a proprietary Predictive Query Language that turns natural-language-style questions into model queries, and reports Kumo connects to data warehouses such as Snowflake and Databricks and automates data cleaning, joins, and feature preparation, claiming a reduction in manual setup effort of up to 95% (a vendor figure). Fortune has cited cofounder Jure Leskovec describing the product as letting a user point the model at their data, define a target such as churn, and receive a prediction almost immediately.
Editorial analysis - technical context
Industry-pattern observations: graph-based representations are a common choice when relationships across records carry signal, because they avoid extensive manual feature joins; pairing them with tight warehouse integrations and a declarative query layer is a frequent way to simplify adoption for analytics teams. The approach contrasts with general-purpose, text-only LLM products by focusing on structured, relational enterprise data.
Context and significance
For practitioners, the acquisition signals continued vendor interest in packaging model capabilities together with data connectors and automation. Fortune frames the deal within a broader pattern of Nvidia acquisitions in recent years. Folding a specialized, connector-first predictive platform into Nvidia's stack illustrates a market for vertical, data-warehouse-native model products alongside foundation-model offerings.
Reported limitations
Full terms were not disclosed: Fortune reports Nvidia declined to comment, and the at-least-$400 million figure comes from The Information. The up-to-95% efficiency claim is reported as a vendor statement about the platform's automation.
What to watch
Open questions include whether Nvidia offers Kumo's technology as a standalone service, embeds it in GPU-accelerated inference products, or surfaces it through partnerships with Snowflake and Databricks; whether existing Kumo customers retain access; and whether the team publishes benchmarks on its graph-based approach.
Key Points
- 1Nvidia acquired enterprise predictive-AI startup Kumo AI for a reported $400 million-plus (The Information), with all three cofounders, including Stanford's Jure Leskovec, joining Nvidia.
- 2Editorial analysis: Kumo's graph neural networks and query DSL encode relational business signals across warehouse data, reducing manual feature engineering for churn-style predictions.
- 3Editorial analysis: packaging models with Snowflake and Databricks connectors and automation reflects demand for warehouse-connected, turn-key predictive products rather than LLM-only offerings.
Scoring Rationale
Nvidia acquiring Kumo AI - a predictive-modeling startup with a recognized graph-learning team, for a reported $400 million-plus - is a notable enterprise-AI consolidation that matters to practitioners productionizing models on warehouse data. It is a significant strategic acquisition by the dominant AI hardware vendor, though smaller than the largest mega-deals, placing it in the upper-notable range.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems